Can a variational quantum circuit add useful modeling capacity after a graph neural network has learned the structure of a 2D material? I explored that question through bandgap prediction on the 2DMatPedia corpus, using crystal graphs and compact physics descriptors rather than treating each material as a flat feature vector.

The project developed across three architectures to isolate quantum contributions. Model C established an amplitude-encoded Coulomb-matrix baseline. Model A introduced graph representations and angle encoding. Model B became the mainline: a periodic-boundary-aware SchNet encoder, physics features, a learned compression layer, and an RX-encoded PennyLane VQC head.

Early runs produced plausible losses while drifting toward the target mean. I tracked that behavior with a variance-ratio diagnostic and audited the full data path. The pipeline corrections included training-fold-only normalization, explicit exclusion of target-correlated features, and correct handling of spatial descriptors (std_x, std_y, std_z), establishing a verifiable training process.

The preserved corrected run reached R² ≈ 0.782 and MAE = 0.439 eV, with a variance ratio around 0.93. This is a strong single-fold result on the retained bandgap configuration, demonstrating that the hybrid model learned sample-specific variation rather than collapsing toward an average prediction.

How can a QKD-enabled optical network provision competing connection requests when quantum channels, wavelengths, and time slots are limited, and key negotiation is degraded by noise and eavesdropping? My MS thesis evaluated this by coupling the QKD protocol layer with the network-allocation layer, treating them as an interdependent system.

At the protocol layer, I built hybrid Qiskit simulations for BB84 and E91 protocols to analyze QBER and secure key yields under noise. At the network layer, I built a Python simulation on the 14-node NSFNET topology, allocating wavelengths, time-slots, and dedicated quantum channels to connection requests based on dynamic security tiers.

I implemented three Quantum Key Resource Allocation (QKRA) policies: Specific Security Level (SSL), Adaptive Strong Security Level (ASSL), and Adaptive Weak Security Level (AWSL). A priority queue processes high-security requests first. ASSL elevates a request when stronger-security resources have capacity, while AWSL lowers the assigned security tier to maintain connection success when the requested pool is saturated. The comparison tracked allocation success, time-slot utilization, and security-weighted provisioning under increasing traffic.

After correcting early indexing logic defects, I analyzed routing and security policies across increasing connection loads. Allocation success stayed near 1.0 under light load and declined predictably as requests exceeded available resources. Time-slot Utilization Ratio (TUR) rose toward saturation, and Network Security Performance (NSP) scaled with provisioned, security-weighted requests. At the protocol layer, escalating noise and eavesdropping validated the transition boundary between Hamming correction and Cascade reconciliation as QBER increased and secure-key yield dropped. In the final MS Thesis, I delivered the complete system architecture and benchmarks.

The linked public repository is a tested, maintainable extraction of the broader thesis. It exposes the full 14-node NSFNET allocation study with priority scheduling, k-shortest paths, first-fit allocation, and performance metrics (SRCR, TUR, NSP), alongside selected BB84 and Hamming utilities. Raw experimental sandboxes, E91 implementations, and Cascade reconciliation experiments remain in the complete thesis archive. This project represents a controlled numerical simulation of QKD network behavior, not a hardware deployment or a production cryptographic stack.

Post-quantum migration is a systems and risk-management problem, not a library swap. For long-retention public-sector data, the difficult question is how to address the $7.1B federal migration scope and reduce harvest-now-decrypt-later exposure while preserving operational continuity.

I authored five-phase migration roadmaps intended for federal agency deployment scenarios, moving from governance and cryptographic discovery through hybrid pilots, infrastructure rollout, and ongoing crypto-agility. The roadmap mapped NIST FIPS 203 (ML-KEM), FIPS 204 (ML-DSA), and FIPS 205 (SLH-DSA) to deployment decisions rather than presenting the standards as an isolated algorithm list.

The framework used CBOM discovery aligned with CycloneDX to expose cryptographic dependencies, then applied Mosca's Theorem to prioritize assets based on the relationship between migration time, required confidentiality shelf-life, and the quantum-threat horizon. It explicitly defined the transitional hybrid pattern (RSA-3072 → ML-KEM-768) to ensure compliance while mitigating immediate exposure.

The most valuable part of the roadmap was its analysis of ten critical deployment loopholes. Instead of assuming a seamless algorithm swap, I mapped the specific failure modes that federal agency deployment scenarios must avoid: HSMs whose external APIs support PQC but whose internal boot firmware still relies on classical ECC, the bandwidth tax of 2.7KB ML-KEM keys on legacy infrastructure, and the reality of crypto-agility anti-patterns hardcoded into third-party supply chains.

This public summary omits agency names, organization-specific architecture, inventory data, and internal timelines. The work demonstrates strategic roadmap and technical-writing capability; it does not claim production deployment or implementation of the PQC primitives themselves.

Power-grid analytics is not one problem: unit commitment, load forecasting, fault classification, and power-flow analysis impose different mathematical and operational constraints. I evaluated quantum methods from those workload structures outward, instead of starting from generic quantum-advantage claims.

I mapped nine algorithm families—including QSVM, QNN, VQELM, QAOA, VQLS, and QSVD—to four grid workload classes. I then screened each mapping against encoding burden, circuit depth, noise sensitivity, classical-baseline strength, and hybrid integration requirements.

The analysis explicitly penalized methods that assume the existence of efficient Quantum RAM (QRAM) for large sensor datasets, distinguishing plausible near-term hybrid candidates (like VQC-based classifiers) from methods whose value depends on fault-tolerant hardware and deep circuits (like HHL for power-flow).

The resulting white paper and decision framework provide a practical basis for deciding what merits deeper validation and what remains premature. The framework treats literature-reported complexity and performance claims as conditional evidence, not as project-produced benchmarks.

This was a literature and feasibility assessment, not an implemented grid optimizer. Client-specific deliverables remain confidential, and this public portfolio omits any project-produced performance benchmarks or quantum-speedup claims.

Can quantum-kernel methods recover secret information from ML-KEM/Kyber power traces using fewer observations than strong classical baselines, or do classical methods remain superior under a fair comparison?

I mapped a staged path: validate the profiling side-channel pipeline on public ASCAD AES traces, gate the transition to Kyber on access to target-specific power traces, and compare reproducible CNN and quantum-kernel key-rank curves under documented feature constraints. ASCAD validates the methodology; it does not stand in for a Kyber result.

The design requires signal-bearing features before quantum-model training, reproducible classical baselines, Kyber-specific data before making Kyber claims, and a Bowles-style classical-kernel control before any quantum-advantage claim. These gates make a negative result as technically useful as a positive one.

I synthesized the relevant literature, selected and presented the research direction, and defined the candidate datasets, model families, success metrics, risks, and go/no-go conditions. The completed output is a technical reference and evaluation plan; it is not a completed attack or key-recovery result.

I retain the framework as an independent research direction, ready for experimental execution once target-specific hardware and high-resolution power-trace data support a defensible comparison.

Quantum adoption is a workforce and operating-model challenge as much as a technology problem. Utilities need distinct capabilities at the field, engineering, and research levels to migrate cryptography, evaluate quantum sensors, and test hybrid computing without treating every employee as a quantum specialist.

I organized the curriculum around technology maturity and operational urgency. Defend focuses on immediate post-quantum cryptography readiness. Sense covers near-term quantum-sensor and resilient-timing pilots. Compute targets longer-term hybrid quantum-classical experiments gated by rigorous comparison against strong classical baselines.

To operationalize the strategy, I designed a three-tier training framework called the National Quantum Grid Academy (QGA). A multi-week Technician Track emphasizes hardware installation and secure data collection. A multi-month Engineer Track covers hybrid workflows, benchmarking, and PQC migration projects. Finally, a research-grade Specialist Track supports advanced algorithm and systems research.

To make the curriculum actionable, I defined a regional delivery model, track-specific capstones, and governance responsibilities. The goal was to tie technical training directly to auditable adoption decisions rather than producing another abstract strategy document.

The project culminated in a white paper and curriculum design proposal delivered to stakeholders. It provides a structured workforce framework but is not an actively deployed academy. This public summary omits client-specific architecture.

Quantum technology reporting is uniquely vulnerable to hype. Quantum Buzzz required editorial judgment capable of distinguishing credible scientific reporting from unsupported novelty claims, inflated performance metrics, and citation drift.

I designed an editorial workflow that decomposed high-risk claims—specifically targeting words like 'breakthrough' or 'proven'—and checked their integrity against primary literature. This involved tracing assertions back to origin DOIs, arXiv preprints, and official benchmarks.

Complex quantum claims passed through a strict verification rubric. The workflow routed drafts toward acceptance, citation repair, overclaim revision, or outright rejection, catching false-novelty narratives and misleading causal leaps before they reached the public.

Beyond individual article reviews, the workflow produced reusable editorial guides, verdict rubrics, and batch fact-check outputs, turning ad-hoc editing into a defensible, source-grounded technical review process.

The public portfolio summarizes the editorial system and its outcomes. Internal editorial rubrics, unpublished drafts, and proprietary review matrices remain confidential.

I investigated pairwise entanglement signatures in a 1D transverse-field Ising chain near its quantum phase transition ($J/h = 1$) under Periodic Boundary Conditions. The goal was to trace out $N-2$ spins from the ground state and calculate Wootters concurrence to locate the inflection point and derivative extremum that mark the phase transition.

To overcome the exponential scaling of the Hilbert space ($k = 2^N$), the codebase evolved through two major phases. The initial exact diagonalization pipeline was memory-bottlenecked at $N \approx 12$ because it required a dense $4096 \times 4096$ complex matrix ($O(2^{2N})$ storage). I refactored the Hamiltonian construction to use Armadillo's sparse representations (`sp_mat`) and offloaded the ground-state search to an ARPACK solver wrapper (`eigs_sym`), shifting to $O(2^N)$ vector storage and extending tractable calculations to $N = 22$ spins.

During testing, compilation dependencies on external libraries (`QIClib`'s partial trace) caused cross-environment compiler mismatches between local workstations and lab cluster setups. To make the solver self-contained, I stripped the third-party dependency and authored a custom native partial trace function to explicitly compute the $4 \times 4$ density matrix directly from the ground-state coefficient vector.

Using the sparse numerical solver, I manually constructed the spin-flipped reduced density matrices $\tilde{\rho}_{12} = (\sigma_y \otimes \sigma_y) \rho_{12}^* (\sigma_y \otimes \sigma_y)$ to calculate Wootters concurrence and its derivative $dC/d\lambda$. The output plots verified critical behavior: the characteristic sharpening of the concurrence inflection and the scaling of the derivative peak exactly at $J/h = 1$, with coordinates correctly shifted using midpoints to align the finite-difference derivatives.

Exact diagonalization and sparse iterative methods still scale exponentially in memory and compute, bounding the simulations to $N \le 22$ on single-machine configurations. This project represents a controlled numerical physics study; it does not claim scalable quantum hardware simulation or tensor-network (DMRG) scaling.